Integrating Large Language Models with ABINIT, OPTIMADE, and jobflow-remote for Automated Materials Simulations

Large Language Models (LLMs) have shown remarkable potential in the computational materials science field, enabling more accessible and automated simulation environments. Building on preliminary work integrating LLMs with Retrieval-Augmented Generation (RAG) frameworks to support user interaction with the Abinit software ecosystem, we now extend this approach toward a broader AI-driven research assistant for atomistic simulations. Specifically, we develop and test an LLM-based translator that converts natural language queries into OPTIMADE-compliant requests, allowing seamless retrieval of crystal structures from multiple materials databases. In parallel, we interface an LLM with jobflow-remote to autonomously execute and manage calculations on remote high-performance computing clusters. Together, these components outline a unified agent capable of retrieving relevant materials data, determining appropriate simulation workflows using Abinit resources through the RAG system, and launching corresponding computations. We discuss this architecture within the Model Context Protocol (MCP) framework as a standardized means of interconnecting these AI-powered tools, paving the way for more automated and adaptive computational materials discovery.